The focus of the policy section is on developing a set of methods that focus both modelers and policymakers on models in a way that makes decision alternatives clearer and improves the power of data and of models to inform policy. The overall strategy is called Inference Robustness Assessment (IRA). It encapsulates basic traditions in science into two iterative processes. The first assesses the validity of inferences made from a particular model form by realistically relaxing simplifying assumptions in that model form and assessing whether greater realism has the potential to change the inferences made when analyzing the model and analyzing data with the model. This iterative loop is pursued when there is enough data to make a definitive inference about policy. When there is not enough data, a second identifiability analysis loop is pursued to determine what additional data will best inform a decision. Identifiability means that data can determine model characteristics uniquely. For example only one policy choice is consistent with the data. The IRA approach does not focus on parameter identifiability as traditional identifiability analysis does. It focuses on making inferences identifiable which may occur even when no parameters are identifiable. A key to making science serve policy using IRA is to begin with simple policy choices of high relevance to the policymakers and with simple models that have face validity for potentially informing that policy choice. We will examine aspects of communication between modelers and policymakers to establish good beginnings leading to successful policymaking. Quite often policy choices entail doing things with limited resources. A key to making models policy relevant is, therefore, incorporating parameters and structures to enable cost effectiveness and cost benefit analyses using the model. Policy choices can be influenced by the chances of different outcomes given different policy decisions. Therefore IRA incorporates a full range of deterministic to stochastic models and employs Bayesian inference to establish the joint ranges of model parameters that are consistent with data. This approach has the advantage that while pursuing the best scientifically supported inference, it is able to inform a policy choice at any level of depth into the IRA process.